Cradl

Cradl AI is a document-processing platform for automation teams. Its current website describes extracting structured data from business documents, splitting and classifying files, and connecting results to Power Automate, Zapier, n8n, APIs or webhooks. Confidence thresholds, cross-model checks and business rules help route uncertain predictions to human review. These controls support a review workflow; they do not establish that every extracted field is correct. The site identifies Lucidtech AS as the supplier in its service terms.

The public pricing page defaults to annual billing. Free includes fifty processed pages per month, one agent and one seat. Builder displays $40 per month, billed annually, with five hundred pages, two agents and two seats. Pro displays $240 per month, billed annually, with five thousand pages, five agents and five seats. Processing a document page again counts again, so repeated runs affect the allowance. Enterprise requires a quote. Currency code, taxes, checkout totals and overage arrangements were not independently confirmed. Retention controls appear on Pro and Enterprise, while enterprise access controls and SSO are separate plan features.

The homepage says customer data is not used to train shared models. This is a supplier statement, not an independent security audit. The controller privacy page covers business contacts and analytics; processing customer documents is governed separately by a data-processing agreement. Those separate terms and retention settings need checking before confidential uploads. The EULA grants service access without ownership of models or APIs, makes customers responsible for input rights, and permits suspension under specified circumstances. End-of-service transfer and deletion have contractual exceptions.

The public signup route displayed an email field, account-creation button, Google and Microsoft options, and links to the service terms and privacy policy. No account, document upload, API request or extraction test was performed; authenticated workflows and extraction quality remain untested.